unrag 0.2.1 → 0.2.3

package/README.md

@@ -10,13 +10,13 @@ It installs small, auditable source files into your repo:
 ## Usage
 
 ```bash
-bunx unrag init
+bunx unrag@latest init
 ```
 
 ### Common flags
 
 ```bash
-bunx unrag init --yes --store drizzle --dir lib/unrag --alias @unrag
+bunx unrag@latest init --yes --store drizzle --dir lib/unrag --alias @unrag
 ```
 
 - `--store`: `drizzle` | `prisma` | `raw-sql`

package/dist/cli/index.js

@@ -74,8 +74,7 @@ var writeText = async (filePath, content) => {
 var renderUnragConfig = (content, selection) => {
   const installImportBase = `./${selection.installDir.replace(/\\/g, "/")}`;
   const baseImports = [
-    `import { createContextEngine, defineConfig } from "${installImportBase}/core";`,
-    `import { createAiEmbeddingProvider } from "${installImportBase}/embedding/ai";`
+    `import { defineUnragConfig } from "${installImportBase}/core";`
   ];
   const storeImports = [];
   const storeCreateLines = [];
@@ -93,24 +92,14 @@ var renderUnragConfig = (content, selection) => {
 `);
   const createEngineBlock = [
     `export function createUnragEngine() {`,
-    `  const embedding = createAiEmbeddingProvider({`,
-    `    model: unragConfig.embedding.model,`,
-    `    timeoutMs: unragConfig.embedding.timeoutMs,`,
-    `  });`,
     ...storeCreateLines,
     ``,
-    `  return createContextEngine(`,
-    `    defineConfig({`,
-    `      embedding,`,
-    `      store,`,
-    `      defaults: unragConfig.chunking,`,
-    `    })`,
-    `  );`,
+    `  return unrag.createEngine({ store });`,
     `}`,
     ``,
     `export async function retrieve(query: string) {`,
     `  const engine = createUnragEngine();`,
-    `  return engine.retrieve({ query, topK: unragConfig.retrieval.topK });`,
+    `  return engine.retrieve({ query, topK: unrag.defaults.retrieval.topK });`,
     `}`
   ].join(`
 `);
@@ -147,6 +136,10 @@ async function copyRegistryFiles(selection) {
       src: path2.join(selection.registryRoot, "core/index.ts"),
       dest: path2.join(installBaseAbs, "core/index.ts")
     },
+    {
+      src: path2.join(selection.registryRoot, "core/assets.ts"),
+      dest: path2.join(installBaseAbs, "core/assets.ts")
+    },
     {
       src: path2.join(selection.registryRoot, "core/types.ts"),
       dest: path2.join(installBaseAbs, "core/types.ts")
@@ -163,6 +156,10 @@ async function copyRegistryFiles(selection) {
       src: path2.join(selection.registryRoot, "core/context-engine.ts"),
       dest: path2.join(installBaseAbs, "core/context-engine.ts")
     },
+    {
+      src: path2.join(selection.registryRoot, "core/delete.ts"),
+      dest: path2.join(installBaseAbs, "core/delete.ts")
+    },
     {
       src: path2.join(selection.registryRoot, "core/ingest.ts"),
       dest: path2.join(installBaseAbs, "core/ingest.ts")
@@ -262,6 +259,70 @@ async function copyConnectorFiles(selection) {
     await writeText(dest, raw);
   }
 }
+async function copyExtractorFiles(selection) {
+  const toAbs = (projectRelative) => path2.join(selection.projectRoot, projectRelative);
+  const installBaseAbs = toAbs(selection.installDir);
+  const extractorRegistryAbs = path2.join(selection.registryRoot, "extractors", selection.extractor);
+  const sharedRegistryAbs = path2.join(selection.registryRoot, "extractors", "_shared");
+  if (!await exists(extractorRegistryAbs)) {
+    throw new Error(`Unknown extractor registry: ${path2.relative(selection.registryRoot, extractorRegistryAbs)}`);
+  }
+  const extractorFiles = await listFilesRecursive(extractorRegistryAbs);
+  const sharedFiles = await exists(sharedRegistryAbs) ? await listFilesRecursive(sharedRegistryAbs) : [];
+  const destRootAbs = path2.join(installBaseAbs, "extractors", selection.extractor);
+  const sharedDestRootAbs = path2.join(installBaseAbs, "extractors", "_shared");
+  const nonInteractive = Boolean(selection.yes) || !process.stdin.isTTY;
+  for (const src of extractorFiles) {
+    if (!await exists(src)) {
+      throw new Error(`Registry file missing: ${src}`);
+    }
+    const rel = path2.relative(extractorRegistryAbs, src);
+    const dest = path2.join(destRootAbs, rel);
+    if (await exists(dest)) {
+      if (nonInteractive) {
+        continue;
+      }
+      const answer = await confirm({
+        message: `Overwrite ${path2.relative(selection.projectRoot, dest)}?`,
+        initialValue: false
+      });
+      if (isCancel(answer)) {
+        cancel("Cancelled.");
+        return;
+      }
+      if (!answer) {
+        continue;
+      }
+    }
+    const raw = await readText(src);
+    await writeText(dest, raw);
+  }
+  for (const src of sharedFiles) {
+    if (!await exists(src)) {
+      throw new Error(`Registry file missing: ${src}`);
+    }
+    const rel = path2.relative(sharedRegistryAbs, src);
+    const dest = path2.join(sharedDestRootAbs, rel);
+    if (await exists(dest)) {
+      if (nonInteractive) {
+        continue;
+      }
+      const answer = await confirm({
+        message: `Overwrite ${path2.relative(selection.projectRoot, dest)}?`,
+        initialValue: false
+      });
+      if (isCancel(answer)) {
+        cancel("Cancelled.");
+        return;
+      }
+      if (!answer) {
+        continue;
+      }
+    }
+    const raw = await readText(src);
+    await writeText(dest, raw);
+  }
+}
 
 // cli/lib/json.ts
 import { readFile as readFile2, writeFile as writeFile2 } from "node:fs/promises";
@@ -347,6 +408,37 @@ function depsForConnector(connector) {
   }
   return { deps, devDeps };
 }
+function depsForExtractor(extractor) {
+  const deps = {};
+  const devDeps = {};
+  if (extractor === "pdf-llm") {
+    deps["ai"] = "^5.0.113";
+  }
+  if (extractor === "pdf-text-layer") {
+    deps["pdfjs-dist"] = "^5.4.149";
+  }
+  if (extractor === "pdf-ocr") {}
+  if (extractor === "image-ocr" || extractor === "image-caption-llm") {
+    deps["ai"] = "^5.0.113";
+  }
+  if (extractor === "audio-transcribe" || extractor === "video-transcribe") {
+    deps["ai"] = "^5.0.113";
+  }
+  if (extractor === "video-frames") {
+    deps["ai"] = "^5.0.113";
+  }
+  if (extractor === "file-text") {}
+  if (extractor === "file-docx") {
+    deps["mammoth"] = "^1.10.0";
+  }
+  if (extractor === "file-pptx") {
+    deps["jszip"] = "^3.10.1";
+  }
+  if (extractor === "file-xlsx") {
+    deps["xlsx"] = "^0.18.5";
+  }
+  return { deps, devDeps };
+}
 function installCmd(pm) {
   if (pm === "bun")
     return "bun install";
@@ -550,7 +642,8 @@ async function initCommand(args) {
     storeAdapter: storeAdapterAnswer,
     aliasBase,
     version: CONFIG_VERSION,
-    connectors: existing?.connectors ?? []
+    connectors: existing?.connectors ?? [],
+    extractors: existing?.extractors ?? []
   };
   await writeJsonFile(path5.join(root, CONFIG_FILE), config);
   const pm = await detectPackageManager(root);
@@ -578,9 +671,34 @@ async function initCommand(args) {
 import { outro as outro2 } from "@clack/prompts";
 import path6 from "node:path";
 import { fileURLToPath as fileURLToPath2 } from "node:url";
+
+// cli/lib/constants.ts
+var UNRAG_SITE_URL = (process.env.UNRAG_SITE_URL ?? process.env.UNRAG_DOCS_BASE_URL)?.trim() || "https://unrag.dev";
+var UNRAG_GITHUB_REPO_URL = "https://github.com/BetterStacks/unrag";
+function docsUrl(siteRelativePath) {
+  const p = siteRelativePath.startsWith("/") ? siteRelativePath : `/${siteRelativePath}`;
+  const base = UNRAG_SITE_URL.endsWith("/") ? UNRAG_SITE_URL : `${UNRAG_SITE_URL}/`;
+  return new URL(p.replace(/^\/+/, "/"), base).toString();
+}
+
+// cli/commands/add.ts
 var CONFIG_FILE2 = "unrag.json";
 var __filename3 = fileURLToPath2(import.meta.url);
 var __dirname3 = path6.dirname(__filename3);
+var AVAILABLE_EXTRACTORS = [
+  "pdf-llm",
+  "pdf-text-layer",
+  "pdf-ocr",
+  "image-ocr",
+  "image-caption-llm",
+  "audio-transcribe",
+  "video-transcribe",
+  "video-frames",
+  "file-text",
+  "file-docx",
+  "file-pptx",
+  "file-xlsx"
+];
 var parseAddArgs = (args) => {
   const out = {};
   for (let i = 0;i < args.length; i++) {
@@ -589,8 +707,17 @@ var parseAddArgs = (args) => {
       out.yes = true;
       continue;
     }
-    if (!out.connector && !a.startsWith("-")) {
-      out.connector = a;
+    if (!out.kind && a && !a.startsWith("-")) {
+      if (a === "extractor") {
+        out.kind = "extractor";
+        continue;
+      }
+      out.kind = "connector";
+      out.name = a;
+      continue;
+    }
+    if (out.kind === "extractor" && !out.name && a && !a.startsWith("-")) {
+      out.name = a;
       continue;
     }
   }
@@ -602,23 +729,24 @@ async function addCommand(args) {
     throw new Error("Could not find a project root (no package.json found).");
   }
   const parsed = parseAddArgs(args);
-  const connector = parsed.connector;
-  if (!connector) {
-    outro2(`Usage: unrag add <connector>
-
-Available connectors: notion`);
-    return;
-  }
-  if (connector !== "notion") {
-    outro2(`Unknown connector: ${connector}
-
-Available connectors: notion`);
+  const kind = parsed.kind ?? "connector";
+  const name = parsed.name;
+  if (!name) {
+    outro2([
+      "Usage:",
+      "  unrag add <connector>",
+      "  unrag add extractor <name>",
+      "",
+      "Available connectors: notion",
+      `Available extractors: ${AVAILABLE_EXTRACTORS.join(", ")}`
+    ].join(`
+`));
     return;
   }
   const configPath = path6.join(root, CONFIG_FILE2);
   const config = await readJsonFile(configPath);
   if (!config?.installDir) {
-    throw new Error(`Missing ${CONFIG_FILE2}. Run \`unrag init\` first.`);
+    throw new Error(`Missing ${CONFIG_FILE2}. Run \`unrag@latest init\` first.`);
   }
   const cliPackageRoot = await findUp(__dirname3, "package.json");
   if (!cliPackageRoot) {
@@ -626,40 +754,120 @@ Available connectors: notion`);
   }
   const registryRoot = path6.join(cliPackageRoot, "registry");
   const nonInteractive = parsed.yes || !process.stdin.isTTY;
-  await copyConnectorFiles({
+  const pkg = await readPackageJson(root);
+  if (kind === "connector") {
+    const connector = name;
+    if (connector !== "notion") {
+      outro2(`Unknown connector: ${name}
+
+Available connectors: notion`);
+      return;
+    }
+    await copyConnectorFiles({
+      projectRoot: root,
+      registryRoot,
+      installDir: config.installDir,
+      connector,
+      yes: nonInteractive
+    });
+    const { deps: deps2, devDeps: devDeps2 } = depsForConnector(connector);
+    const merged2 = mergeDeps(pkg, deps2, devDeps2);
+    if (merged2.changes.length > 0) {
+      await writePackageJson(root, merged2.pkg);
+    }
+    const connectors = Array.from(new Set([...config.connectors ?? [], connector])).sort();
+    await writeJsonFile(configPath, { ...config, connectors });
+    outro2([
+      `Installed connector: ${connector}.`,
+      "",
+      `- Code: ${path6.join(config.installDir, "connectors", connector)}`,
+      `- Docs: ${docsUrl(`/docs/connectors/${connector}`)}`,
+      "",
+      merged2.changes.length > 0 ? `Added deps: ${merged2.changes.map((c) => c.name).join(", ")}` : "Added deps: none",
+      nonInteractive ? "" : "Tip: keep NOTION_TOKEN server-side only (env var)."
+    ].filter(Boolean).join(`
+`));
+    return;
+  }
+  const extractor = name;
+  if (!extractor || !AVAILABLE_EXTRACTORS.includes(extractor)) {
+    outro2(`Unknown extractor: ${name}
+
+Available extractors: ${AVAILABLE_EXTRACTORS.join(", ")}`);
+    return;
+  }
+  await copyExtractorFiles({
     projectRoot: root,
     registryRoot,
     installDir: config.installDir,
-    connector,
+    extractor,
     yes: nonInteractive
   });
-  const pkg = await readPackageJson(root);
-  const { deps, devDeps } = depsForConnector(connector);
+  const { deps, devDeps } = depsForExtractor(extractor);
   const merged = mergeDeps(pkg, deps, devDeps);
   if (merged.changes.length > 0) {
     await writePackageJson(root, merged.pkg);
   }
-  const connectors = Array.from(new Set([...config.connectors ?? [], connector])).sort();
-  await writeJsonFile(configPath, { ...config, connectors });
+  const extractors = Array.from(new Set([...config.extractors ?? [], extractor])).sort();
+  await writeJsonFile(configPath, { ...config, extractors });
   outro2([
-    `Installed connector: ${connector}.`,
+    `Installed extractor: ${extractor}.`,
     "",
-    `- Code: ${path6.join(config.installDir, "connectors", connector)}`,
-    `- Docs: /docs/connectors/${connector}`,
+    `- Code: ${path6.join(config.installDir, "extractors", extractor)}`,
     "",
     merged.changes.length > 0 ? `Added deps: ${merged.changes.map((c) => c.name).join(", ")}` : "Added deps: none",
-    nonInteractive ? "" : "Tip: keep NOTION_TOKEN server-side only (env var)."
+    "",
+    `Next: import the extractor and pass it to createContextEngine({ extractors: [...] }).`
   ].filter(Boolean).join(`
 `));
 }
 
 // cli/run.ts
+function renderHelp() {
+  return [
+    "unrag — vendor-in RAG primitives (ingest/retrieve + adapters) into your repo.",
+    "",
+    "Usage:",
+    "  bunx unrag <command> [options]",
+    "  npx  unrag <command> [options]",
+    "",
+    "Commands:",
+    "  init                Install core files (config + store adapter templates)",
+    "  add <connector>     Install a connector (currently: notion)",
+    "  help                Show this help",
+    "",
+    "Global options:",
+    "  -h, --help           Show help",
+    "  -y, --yes            Non-interactive; accept defaults",
+    "",
+    "init options:",
+    "  --store <adapter>    drizzle | prisma | raw-sql",
+    "  --dir <path>         Install directory (alias: --install-dir)",
+    "  --alias <@name>      Import alias base (e.g. @unrag)",
+    "",
+    "Examples:",
+    "  bunx unrag@latest init",
+    "  bunx unrag@latest init --yes --store drizzle --dir lib/unrag --alias @unrag",
+    "  bunx unrag add notion --yes",
+    "",
+    "Docs:",
+    `  - Quickstart: ${docsUrl("/docs/getting-started/quickstart")}`,
+    `  - CLI:       ${docsUrl("/docs/reference/cli")}`,
+    `  - Notion:    ${docsUrl("/docs/connectors/notion")}`,
+    "",
+    "Repo:",
+    `  ${UNRAG_GITHUB_REPO_URL}`,
+    "",
+    "Tip:",
+    "  After `init`, open the generated unrag.md for schema + env vars (DATABASE_URL)."
+  ].join(`
+`);
+}
 async function run(argv) {
   const [, , command, ...rest] = argv;
   intro("unrag");
   if (!command || command === "help" || command === "--help" || command === "-h") {
-    outro3(["Usage:", "", "- unrag init", "- unrag add <connector>"].join(`
-`));
+    outro3(renderHelp());
     return;
   }
   if (command === "init") {
@@ -670,7 +878,8 @@ async function run(argv) {
     await addCommand(rest);
     return;
   }
-  outro3(`Unknown command: ${command}`);
+  outro3([`Unknown command: ${command}`, "", renderHelp()].join(`
+`));
   process.exitCode = 1;
 }
 

package/package.json

@@ -1,10 +1,11 @@
 {
   "name": "unrag",
   "type": "module",
+  "repository": "https://github.com/BetterStacks/unrag",
   "bin": {
     "unrag": "./dist/cli/index.js"
   },
-  "version": "0.2.1",
+  "version": "0.2.3",
   "private": false,
   "license": "Apache-2.0",
   "devDependencies": {

package/registry/config/unrag.config.ts

@@ -2,10 +2,10 @@
  * Root Unrag config (generated).
  *
  * This file is meant to be the single place you tweak:
+ * - Defaults (chunking + retrieval)
+ * - Engine settings (storage, asset processing, extractors)
  * - Embedding provider/model/timeouts
- * - Chunking defaults
- * - Retrieval defaults
- * - How you construct your DB client (Pool/Prisma/etc)
+ * - How you construct your DB client (Pool/Prisma/etc) and vector store adapter
  *
  * The files under your install dir (e.g. `lib/unrag/**`) are intended to be
  * treated like vendored source code.
@@ -13,7 +13,8 @@
 
 // __UNRAG_IMPORTS__
 
-export const unragConfig = {
+export const unrag = defineUnragConfig({
+  defaults: {
   chunking: {
     chunkSize: 200,
     chunkOverlap: 40,
@@ -21,11 +22,143 @@ export const unragConfig = {
   retrieval: {
     topK: 8,
   },
+  },
   embedding: {
-    model: "openai/text-embedding-3-small",
-    timeoutMs: 15_000,
+    provider: "ai",
+    config: {
+      type: "text",
+      model: "openai/text-embedding-3-small",
+      timeoutMs: 15_000,
+    },
+  },
+  engine: {
+  /**
+   * Storage controls.
+   *
+   * - storeChunkContent: whether `chunk.content` is persisted and returned by retrieval.
+   * - storeDocumentContent: whether the full original document text is stored in `documents.content`.
+   */
+  storage: {
+    storeChunkContent: true,
+    storeDocumentContent: true,
+  },
+    /**
+     * Optional extractor modules that can process non-text assets into text outputs.
+     *
+     * To install:
+     * - `unrag add extractor pdf-llm`
+     *
+     * Then import it in this file and add it here, for example:
+     * - `import { createPdfLlmExtractor } from "./lib/unrag/extractors/pdf-llm";`
+     * - `extractors: [createPdfLlmExtractor()]`
+     */
+    extractors: [],
+  /**
+   * Rich media processing controls.
+   *
+   * Notes:
+   * - The library defaults are cost-safe (PDF LLM extraction is off).
+   * - This generated config opts you into PDF extraction for convenience.
+   * - Tighten fetch allowlists/limits in production if you ingest URL-based assets.
+   */
+  assetProcessing: {
+    onUnsupportedAsset: "skip",
+    onError: "skip",
+    concurrency: 4,
+    fetch: {
+      enabled: true,
+      maxBytes: 15 * 1024 * 1024,
+      timeoutMs: 20_000,
+      // allowedHosts: ["..."], // recommended to mitigate SSRF
+    },
+    pdf: {
+      // Fast/cheap text-layer extraction (requires installing a PDF text-layer extractor module).
+      textLayer: {
+        enabled: false,
+        maxBytes: 15 * 1024 * 1024,
+        maxOutputChars: 200_000,
+        minChars: 200,
+        // maxPages: 200,
+      },
+      llmExtraction: {
+        enabled: true,
+        model: "google/gemini-2.0-flash",
+        prompt:
+          "Extract all readable text from this PDF as faithfully as possible. Preserve structure with headings and lists when obvious. Output plain text or markdown only. Do not add commentary.",
+        timeoutMs: 60_000,
+        maxBytes: 15 * 1024 * 1024,
+        maxOutputChars: 200_000,
+      },
+      // Worker-only OCR pipelines typically require native binaries (poppler/tesseract) or external services.
+      ocr: {
+        enabled: false,
+        maxBytes: 15 * 1024 * 1024,
+        maxOutputChars: 200_000,
+        minChars: 200,
+        // maxPages: 200,
+        // pdftoppmPath: "/usr/bin/pdftoppm",
+        // tesseractPath: "/usr/bin/tesseract",
+        // dpi: 200,
+        // lang: "eng",
+      },
+    },
+    image: {
+      ocr: {
+        enabled: false,
+        model: "google/gemini-2.0-flash",
+        prompt:
+          "Extract all readable text from this image as faithfully as possible. Output plain text only. Do not add commentary.",
+        timeoutMs: 60_000,
+        maxBytes: 10 * 1024 * 1024,
+        maxOutputChars: 50_000,
+      },
+      captionLlm: {
+        enabled: false,
+        model: "google/gemini-2.0-flash",
+        prompt:
+          "Write a concise, information-dense caption for this image. Include names, numbers, and labels if visible. Output plain text only.",
+        timeoutMs: 60_000,
+        maxBytes: 10 * 1024 * 1024,
+        maxOutputChars: 10_000,
+      },
+    },
+    audio: {
+      transcription: {
+        enabled: false,
+        model: "openai/whisper-1",
+        timeoutMs: 120_000,
+        maxBytes: 25 * 1024 * 1024,
+      },
+    },
+    video: {
+      transcription: {
+        enabled: false,
+        model: "openai/whisper-1",
+        timeoutMs: 120_000,
+        maxBytes: 50 * 1024 * 1024,
+      },
+      frames: {
+        enabled: false,
+        sampleFps: 0.2,
+        maxFrames: 50,
+        // ffmpegPath: "/usr/bin/ffmpeg",
+        maxBytes: 50 * 1024 * 1024,
+        model: "google/gemini-2.0-flash",
+        prompt:
+          "Extract all readable text from this video frame as faithfully as possible. Output plain text only. Do not add commentary.",
+        timeoutMs: 60_000,
+        maxOutputChars: 50_000,
+      },
+    },
+    file: {
+      text: { enabled: false, maxBytes: 5 * 1024 * 1024, maxOutputChars: 200_000, minChars: 50 },
+      docx: { enabled: false, maxBytes: 15 * 1024 * 1024, maxOutputChars: 200_000, minChars: 50 },
+      pptx: { enabled: false, maxBytes: 30 * 1024 * 1024, maxOutputChars: 200_000, minChars: 50 },
+      xlsx: { enabled: false, maxBytes: 30 * 1024 * 1024, maxOutputChars: 200_000, minChars: 50 },
+    },
+  },
   },
-} as const;
+} as const);
 
 // __UNRAG_CREATE_ENGINE__
 

package/registry/connectors/notion/render.ts

@@ -1,3 +1,5 @@
+import type { AssetInput, AssetKind, Metadata } from "../../core";
+
 type RichText = { plain_text?: string };
 
 export type NotionBlock = {
@@ -20,6 +22,82 @@ const rt = (value: unknown): string => {
 
 const indent = (n: number) => (n > 0 ? "  ".repeat(n) : "");
 
+const asString = (v: unknown) => String(v ?? "").trim();
+
+const supportedAssetKinds = new Set<AssetKind>([
+  "image",
+  "pdf",
+  "audio",
+  "video",
+  "file",
+]);
+
+const toAssetKind = (notionType: string): AssetKind | null => {
+  const t = notionType as AssetKind;
+  return supportedAssetKinds.has(t) ? t : null;
+};
+
+const pickUrl = (payload: any): string | undefined => {
+  const type = String(payload?.type ?? "");
+  if (type === "external") return asString(payload?.external?.url);
+  if (type === "file") return asString(payload?.file?.url);
+  return undefined;
+};
+
+const pickCaption = (payload: any): string => {
+  // Notion captions are typically an array of rich text items.
+  return rt(payload?.caption);
+};
+
+const inferMediaType = (assetKind: AssetKind, payload: any): string | undefined => {
+  if (assetKind === "pdf") return "application/pdf";
+  // Notion does not consistently include media types; keep it optional.
+  return asString(payload?.media_type) || undefined;
+};
+
+const asMetadata = (obj: Record<string, unknown>): Metadata => obj as any;
+
+export function extractNotionAssets(
+  nodes: NotionBlockNode[],
+  opts: { maxDepth?: number } = {}
+): AssetInput[] {
+  const maxDepth = opts.maxDepth ?? 6;
+  const out: AssetInput[] = [];
+
+  const walk = (node: NotionBlockNode, depth: number) => {
+    if (depth > maxDepth) return;
+    const b = node.block as any;
+    const kind = toAssetKind(String(b.type ?? ""));
+    if (kind) {
+      const payload = b[kind];
+      const url = pickUrl(payload);
+      if (url) {
+        const caption = pickCaption(payload).trim();
+        const mediaType = inferMediaType(kind, payload);
+        out.push({
+          assetId: String(b.id),
+          kind,
+          data: { kind: "url", url, ...(mediaType ? { mediaType } : {}) },
+          uri: url,
+          ...(caption ? { text: caption } : {}),
+          metadata: asMetadata({
+            connector: "notion",
+            notionBlockId: String(b.id),
+            notionBlockType: String(b.type),
+          }),
+        });
+      }
+    }
+
+    for (const child of node.children) {
+      walk(child, depth + 1);
+    }
+  };
+
+  for (const n of nodes) walk(n, 0);
+  return out;
+}
+
 export function renderNotionBlocksToText(
   nodes: NotionBlockNode[],
   opts: { maxDepth?: number } = {}